Improved Industrial Robot Positional Accuracy for Machining with Bias Correction

Robotic machining has the potential to provide advantages as a substitute for conventional CNC machine tool operations. However, conventional industrial robots are restricted to low accuracy tasks due to their poor positional accuracy. This creates challenges in achieving the tolerances required for machining tasks. Data-based modelling of the positional error data is a potential solution which learns the positional errors in order to compensate and minimise them. There has been some success in improving industrial robot accuracy in research literature, by first calibrating the kinematic model and then using machine learning (ML)-based bias correction to learn the positional errors. However, the limitations of ML-based bias correction applied to the industrial robot positional accuracy problem have not been fully explored with the accuracies required to achieve tight machining tolerances. Mapping the positional errors with a greater resolution of training data, and reducing the burden on bias correction by calibrating the kinematic model with a higher level of calibration, are two examples which have the potential to improve accuracy. This thesis focusses on both training data resolution and bias reduction to maximise outcomes whilst informing trade-offs when using ML-based bias correction in this application. The key finding of this thesis is that substantial gains in accuracy can be achieved using ML-based bias correction and that the accuracy limit can be achieved with practicable amounts of data gathering and processing. Also that calibration prior to bias correction did not significantly improve overall accuracy for the cases investigated. This suggests that data may be better utilised in training the bias corrector rather than for calibration of the physical model. In conclusion, ML-based bias correction methods can provide a solution that provides substantial gains in positional accuracy for conventional industrial robots, bringing them to a level that may facilitate broader adoption in machining applications.